Table of Contents
What's also worth knowing
Before we look at a structured way to think about note taking, it’s worth stopping one more time at some meaningful angles we need to consider:
- Which difficulties tend to be most affected by AI usage?
- What can a client notice themselves about their own usage?
- OCD: AI designed to agree with or reassure can feed a reassurance-seeking compulsion
- Psychosis: AI may trigger or worsen delusional beliefs, including in people with no prior history. This has sometimes been described informally as “AI psychosis”
- Mania and hypomania: AI can feed into heightened emotional states or reinforce beliefs and behaviours that make someone more unwell
- Health anxiety: using AI to check symptoms can become compulsive, and the information returned may simply be wrong
- Eating problems: AI can give harmful diet or food information when asked
- Depression: for some clients, the volume of AI in daily life is itself a low-mood driver, distinct from their direct use of it
It’s important to remember these more specific difficulties when using the below tools to document or safety plan a client’s use of AI. Similarly, alongside what a clinician might pick up on from the outside, it’s worth knowing what a client might notice about their own use if they stopped to check:
- Spending longer and longer periods on AI, or reaching for it as the first response to anxiety or boredom
- Repeating the same conversations with it (constantly seeking reassurance)
- Feeling worse after using it, or anxious at the thought of not having access to it
- A shift from practical use (work, admin) toward increasingly emotional or personal use
- Not trusting their own judgement without checking it first
- AI use visibly affecting sleep, relationships, or daily functioning
None of the above immediately mean that you need to be worried about a client. A client could recognise two or three of these in themselves and still be using AI in a way that’s broadly working for them. On the other hand, it could be that someone doesn’t recognise any of them in their own usage and it still might be worth a closer look for other reasons entirely. As in the rest of the course, these are prompts for a conversation, not a checklist or a hard structure to base your decisions on.
So let’s look at the final step now: a structured way to actually formulate and document what’s in front of you with a specific client.
The formulation grid
We know that not every single discussion needs a full detailed note, but it can be useful to structure your thinking and slowing down by going through a few structured questions. The tool below will help you do this. It isn’t meant to be a protocol but more of a discussion prompt to be used alongside your own clinical judgement and the supervision you receive.
The tool walks you through the five questions one at a time rather than all at once. Each question comes with a couple of follow-up prompts so that you can check your first answer and be nudged slightly deeper in the thinking around it. Once you’ve been through all five, there’s a blank version of the same questions waiting for you if you would like to have the same structure in your own notes.
The NHS "P's" model
NHS England’s 2025 Staying Safe from Suicide guidance moved practice away from risk labels and stratification (low, medium, high) toward making a joint safety formulation together with your client based around the following 5 “Ps”:
- Presenting problem
- Precipitating factor
- Protective factor
- Predisposing factor
- Perpetuating factor
When client uses AI, we can map that directly into that structure rather then needing to create a whole new framework to think about it. The below tool does exactly that, one P at a time. You will be guided through with additional thought prompts to make sure you think about it from specific angles.
It’s worth using this alongside the formulation grid above rather than instead of it: the grid asks what the AI use itself is doing for a client, this asks where it sits within the wider safety picture you’re likely already building using your service’s existing approach. And as with the other tool, if you would like a blank version that you can just copy into your own notes, that’s waiting for you at the end.
What to document and where?
| Location | What goes here |
|---|---|
| Safety planning | Anything that changes the shared understanding of safety should be document here. You can use the 5P model here to help structure that type of note but it’s important to make sure you follow in service guidance too. |
| General session notes | This is where you can document your conversations with the client about AI usage. This will generally be as well where routine AI use without a big impact is documented. This could be for example where a client is using AI for psycho-education purposes. |
| Care plan | When a client is actively using AI as a coping strategy between sessions, then it should be documented in their care plan. |
Consent and Sharing Information
There might be times where discussing client’s AI use becomes a question of needing to share information with someone outside of the immediate session. In these cases, there is no need to create a formal AI-specific consent form, because this would fall under your normal information governance policies and legal bases that you already use in your normal safety practices:
- Informed consent
- Risk of serious harm
- Lack of capacity under the Mental Capacity Act
- Child safeguarding under the Children Act
- Another Mental Health Act duty.
It’s important to follow these as well as other local services practices and policies to guide your steps when you feel you need to share information outside of the session.
Reflective Question

A client mentions they’ve been “talking things through” with an AI app most evenings for the past month. They seem calmer in session, but you notice they’ve stopped bringing up their relationship with their sister, which used to come up every week. Work through this example with the formulation grid. What stood out and what would you want to explore further, maybe even with a colleague or in supervision?


